mdadm unmark drive as spare?

gladbach · n00b Joined: 15 Jun 2002 Posts: 50

I have a 4 drive raid 5 array using mdadm. sda1 sdb1 sdc1 sdd1

sdb1 is actually a failed drive, and needs to be replaced. However, somewhere along the way, sdc1 was marked as spare, and I can not seem to get it to go back to active. sdc1 seems completely ok, other than it is marked as spare, and md0 won't run because there are 2 active, 1 spare drives.

is there a way to correct this, or am I screwed?

gladbach · n00b Joined: 15 Jun 2002 Posts: 50

no ideas?

gladbach · n00b Joined: 15 Jun 2002 Posts: 50

I resolved the issue on my own. Once I had a replacement for the physical bad drive, I was able to force the rebuild (previously would error out) using the command

fryfrog · n00b Joined: 09 May 2002 Posts: 45

I just wanted to say thanks for posting this.

I just used this to save my friends 5x1.5T raid5 array.

During a motherboard issue and upgrade, his array failed out a disk and then soon after another disk. At some point during his troubleshooting, the disk that was still "good" was re-added but as a spare. So the array had 3/5 disks and a spare. The 5th disk looked legitimately toasted, the raid superblock was missing and the partition information was odd. It was *probably* fixable, but I doubt it would have been easy.

Using the 3 good disks, I figured out their original positions from 'mdadm -E /dev/sdX' and ended up with something like 'mdadm --create /dev/md0 --assume-clean --level=5 --chunk=64 --raid-devices=5 missing /dev/sdb1 missing /dev/sdd1 /dev/sdf1' which obviously wouldn't work. I randomly put the disk that was "good" (but believed itself to be a spare) in the first "missing" spot, used 'mdadm -o /dev/md0' to mark it read only, but attempting to mount it did not work. I then re-did the command with the "spare" disk in the second missing spot, 'mdadm -o /dev/md0' to make it read only again and mount believed there was an xfs filesystem there (but wouldn't mount it due to a read only block device)!

Amazing!

The superblock of the xfs file system was missing or damaged, so it wouldn't mount to play back the log. A run of 'xfs_repair -L /dev/md0' nuked the log but recovered the file system. The only victim was a single jpg ending up in lost+found.

I hope this helps someone else some day too.

I think the most important things to note are the raid level, chunk and number of devices. Hopefully you know most or all of this, but 'mdadm -E /dev/sd??' should be able to help you figure things out. It will also tell you the position the disk was in the last time it was in the array and *confusingly* also the device it was. During my first attempts, one of the disks said it was in a certain position and called /dev/sde1, so I used that device in that position... but of course they had been re-arranged. I was glad when I realized this.

Note: I used obviously sequential /dev/sdX1 devices in the example, but in real life they were all mixed up because really, who cares where drives are plugged in ;)